![]() |
MeLOn
|
Typedefs | |
using | X = lhsdesign(nX,DX) |
Functions | |
Training | function (Levenberg Marquardt) ffNet |
View network | view (ffNet) %% Save ANN parameters in csv-files % We write csv-files that is read by our MAiNGO model. sNetwork2CSV(ffNet) |
Compute GP predictions in | Matlab (just for information) x_Test_Point |
Plot generated | data (just for information) % Generate a mesh on the inputs[x_1_prediction |
x_2_prediction (i, j)] | |
y_prediction (i, j) | |
Variables | |
file example_training_of_GP m brief Illustrative training script for feedforward artificial neural network in | Matlab |
file example_training_of_GP m brief Illustrative training script for feedforward artificial neural network in RWTH Aachen University | n |
Add path for Gaussian process export functions General | nX = 200 |
Number of training data points | DX = 2 |
Input dimension of data GP | lb = [-3, -3] |
Define Lower bound of inputs | ub = [ 3, 3] |
Define upper bound of inputs | test_func |
Scale inputs onto interval[lb, ub] | Y = cellfun(test_func, num2cell(X,2)) |
Evaluate test_func for all X Tranining of ANNs | numNeurons = [10,8] |
trainFcn = 'trainlm' | |
ffNet = configure(ffNet, X',Y') | |
Transfer | function |
Transfer | purelin |
Transfer ffNet | layers {1}.transferFcn = 'tansig' |
Output Layer ffNet | name = 'myTestANN' |
Devide data set into | training |
Devide data set into | validation |
Devide data set into and test ffNet divideParam | trainRatio = 70/100 |
ffNet divideParam | valRatio = 15/100 |
ffNet divideParam | testRatio = 15/100 |
ffNet | divideFcn = 'divideblock' |
ffNet performParam | normalization = 'standard' |
Training Parameters ffNet trainParam | epochs = 50000 |
ffNet trainParam | max_fail = 6 |
ffNet trainParam | min_grad = 1e-07 |
ffNet trainParam | mu_max = 1e10 |
prediction = ffNet(x_Test_Point) | |
perf = perform(ffNet,prediction,Y) | |
Plot generated | x_2_prediction = meshgrid(linspace(lb(1), ub(1), 20), linspace(lb(2), ub(2),20) ) |
y_prediction = zeros(size(x_2_prediction,1),1) | |
y_std = zeros(size(x_2_prediction,1),1) | |
Evaluate prediction at all mesh points for | i |
Plot generated data | ( | just for | information | ) |
Training function | ( | Levenberg | Marquardt | ) |
Compute GP predictions in Matlab | ( | just for | information | ) |
View network view | ( | ffNet | ) |
ffNet divideFcn = 'divideblock' |
if DX = 2 |
Training Parameters ffNet trainParam epochs = 50000 |
for i |
ffNet layers {1}.transferFcn = 'tansig' |
file example_training_of_GP m brief Illustrative training script for feedforward artificial neural network in Matlab |
ffNet trainParam max_fail = 6 |
ffNet trainParam min_grad = 1e-07 |
ffNet trainParam mu_max = 1e10 |
Output Layer ffNet name = 'myTestANN' |
ffNet performParam normalization = 'standard' |
Add path for Gaussian process export functions General nX = 200 |
perf = perform(ffNet,prediction,Y) |
prediction = ffNet(x_Test_Point) |
Transfer purelin |
Define upper bound of inputs test_func |
ffNet divideParam testRatio = 15/100 |
trainFcn = 'trainlm' |
Devide data set into training |
Define Lower bound of inputs ub = [ 3, 3] |
Devide data set into validation |
ffNet divideParam valRatio = 15/100 |
y_prediction = zeros(size(x_2_prediction,1),1) |
y_std = zeros(size(x_2_prediction,1),1) |